Info 64MB V-Cache on 5XXX Zen3 Average +15% in Games

[Kedas]

Well we know now how they will bridge the long wait to Zen4 on AM5 Q4 2022.
Production start for V-cache is end this year so too early for Zen4 so this is certainly coming to AM4.
+15% Lisa said is "like an entire architectural generation"

 

[Joe NYC]

Why? They are beating the 12900k at 4.5 Ghz,. Even the worst binned 5800X3D will be the Gaming King when compared to Alder Lake

The product if excellent. If the worst binned product can beat the greatest Intel can produced at insane power levels for desktop of near 250 Watts.

The timing is what makes it irrelevant.

Intel threw everything at Alder Lake to make it come first across finish line. Almost 3 months later, AMD mails it in, saying sometimes in Spring reminds me of Fielding Mellish (Woody Allen) in Bananas:

 

[DrMrLordX]

Maybe Milan-X demand caught them by surprise,

Didn't someone post on these forums a few months ago about additional parties showing interest in Milan-X? Might have even been in this thread.

 

[Hitman928]

Didn't someone post on these forums a few months ago about additional parties showing interest in Milan-X? Might have even been in this thread.

Given the early testing results AMD had access to, I just can't imagine AMD wouldn't have known for quite a while that a whole lot of people would want to buy Milan-X. Far more than they could supply.

 

[nicalandia]

They may beat the 12900k but they aren't beating the 12900ks or kf or extreme what ever intel calls there reply release...

Do you think a 0.3 Ghz will even things out? No a chance

 

[Timorous]

Are you saying that 5900x and 5950x bins are less leaky than a 5800x? I'm not 100% sure that's the case given that they have to boost higher than a 5800x in low threaded loads.

How else do you get 12/16c and 2x the cache in the same TDP envelope?

 

[DrMrLordX]

Given the early testing results AMD had access to, I just can't imagine AMD wouldn't have known for quite a while that a whole lot of people would want to buy Milan-X. Far more than they could supply.

Then they never should have let any of their people confirm Zen3D in interviews last year . . .

 

[Abwx]

As it's their best performing gaming processor before. Also gaming performance increased much more than what they informed before, they said 15% @ iso clocks, now it's 15% average improvement with regressed clock and less cores - improvement at iso clocks have to be way over 20%.

It s because the 5800X has half the 5950X s L3 cache, so it benefit more from the augmented L3.

Other than this theyll surely release 5900X/5950X versions if there s a big demand, the 5800X3D is some kind of sales test vehicle.

 

[Hitman928]

How else do you get 12/16c and 2x the cache in the same TDP envelope?

By lowering the all core clocks for higher core count SKUs. . .

 

[nicalandia]

The product if excellent. If the worst binned product can beat the greatest Intel can produced at insane power levels for desktop of near 250 Watts.

The timing is what makes it irrelevant.

Intel threw everything at Alder Lake to make it come first across finish line. Almost 3 months later, AMD mails it in.

It's just a place holder for the "Best Gaming CPU" for the time being just before Zen4 will just clubber stomp anything(including the 5800X3D) that has been released, they Demoed a ES with All Core 5Ghz for crying out loud.

 

[Hitman928]

Then they never should have let any of their people confirm Zen3D in interviews last year . . .

I'm assuming you mean Ryzen3d. If so, that is why I don't buy the argument about unexpected MilanX demand being the catalyst for why we only get the 5800x3d given that they would have known all this back then. I'm not saying Milan-X demand isn't driving this decision, I think it is, only that I don't think there's any way they didn't foresee that demand from day 1 (of the public info coming out). I could be wrong, it's just hard for me to think AMD would be that naive on the subject.

 

[Timorous]

By lowering the all core clocks for higher core count SKUs. . .

Something they do for Milan-X parts.

4.5Ghz max turbo makes it an easier bin than the 5600X.

The part is a reclaim slot for the dregs that don't hit Milan-X bins.

Edit: I presume the single SKU + Spring release is due to not many dies being that bad so it takes a while to build enough inventory without impacting Milan-X supply.

 

[Joe NYC]

I think you were the only once expecting V-cache in Q4 that I have seen and everyone else said it will be a 2022 product and they were right. That AMD would deliver a small number to key server partners / hyperscalers before that was never in question as they are always delivered silicon before it is made available to the broader market as they bring the product to HVM. That's not what we refer to as 'available' though and never have. No one argues that Milan was launched before Ryzen 5000 series even though hyperscalers had it for months before the first Zen3 CPU was for sale.

Microsoft would have to be lying when they said (in early Q4) they are upgrading their entire HPC Azure instance to Milan-X. Why would they lie to their users if they had no product, or set unrealistic expectations? Why would they be advertising Milan-X in Q4 of 2021 if they had no product?

The answer is very simple. Zen3d dies shipped in Q4 as Milan-X, instead of desktop RyZen3d.

Microsoft arrangement must have greatly exceeded any possible returns from desktop sales. Was worth more than the temporary loss of AMD brand value for gamers...

 

[Hitman928]

Something they do for Milan-X parts.

4.5Ghz max turbo makes it an easier bin than the 5600X.

The part is a reclaim slot for the dregs that don't hit Milan-X bins.

Server CPUs typically take lower leakage (and thus lower max turbo) bins. If it gets rejected from Milan-X then it is usually because it is too leaky. Those too leaky CPUs shouldn't have a need for a reduced single core turbo though, only when lots of cores are loaded.

Maybe Milan-X bins are even tighter than Milan to try and get the most efficient dies to somewhat offset the increased power brought by the larger L3? Even if so, the CCD fabrication is separate from the V-cache so if they have CCDs that don't meat stricter Milan-X bins but still can be used for Milan, then why send them for 3d packaging in the first place? To this effect, there shouldn't be any Milan-X cast offs as the CCD will fail binning for Milan-X before being thinned and stacked. Maybe they aren't binning at all prior to stacking but that would seem very foolish and wasteful.

I think AMD's explanation makes the most sense. With the increased cost to produce, they are using the 5800x3d as a test case to see how the market reacts. This will have minimal disruption to their other lines and give AMD market feedback on 3d cache consumer products.

 

[Hitman928]

Microsoft would have to be lying when they said (in early Q4) they are upgrading their entire HPC Azure instance to Milan-X. Why would they lie to their users if they had no product, or set unrealistic expectations? Why would they be advertising Milan-X in Q4 of 2021 if they had no product?

The answer is very simple. Zen3d dies shipped in Q4 as Milan-X, instead of desktop RyZen3d.

Microsoft arrangement must have greatly exceeded any possible returns from desktop sales. Was worth more than the temporary loss of AMD brand value for gamers...

They are upgrading all of their Milan powered HPv3 Azure instances to Milan-X 'soon'. These instances are not yet publicly available and are only available in 1 region under a preview program which you have to be accepted into.

In other words, AMD shipped some early silicon to key partners in 2H21. Most likely very early ES samples were sent even before that for validation purposes. Those partners started to bring up test systems towards the end of the year and Microsoft decided to use some of those test systems (after internal testing) in their preview program for essentially public beta testing (see their preview program). As long as all goes well, they will then fully launch in every region as they receive addition silicon from AMD as AMD ramps HVM in Q122.

This is the expected timeline by pretty much everyone except you for some reason who thought AMD would ever consider shipping early silicon straight to gamers before their business partners, bucking established production timelines set in place years now.

 

[Timorous]

Those too leaky CPUs shouldn't have a need for a reduced single core turbo though, only when lots of cores are loaded.

And which Ryzen class CPU does that fit the bill for? 5800X by any chance since 5900X and 5950X are both less leaky than 5800X.

When it comes to reject Milan-X dies the 5800X is probably the sweet spot for margin and bin tolerance.

 

[Hitman928]

And which Ryzen class CPU does that fit the bill for? 5800X by any chance since 5900X and 5950X are both less leaky than 5800X.

When it comes to reject Milan-X dies the 5800X is probably the sweet spot for margin and bin tolerance.

Are you assuming 59(5)0x are less leaky on average or do you have data on this? This of course is also more complex to answer as there are actually 2 CCDs versus one in this comparison.

 

[Insert_Nickname]

I would do side grade from a 5800X if the resale value is enough. That boost in performance could mean skipping Zen 4 and going Zen 5.

Doesn't seem a half bad upgrade if you can do that.

I think I'll be sitting this one out though, and I've had enough first launch with original 1st gen Zen. I'll wait for second gen AM5 boards. With Zen 4 or 5. Hopefully things in the DDR5 hardware department have settled down by then.

 

[Makaveli]

Doesn't seem a half bad upgrade if you can do that.

I think I'll be sitting this one out though, and I've had enough first launch with original 1st gen Zen. I'll wait for second gen AM5 boards. With Zen 4 or 5. Hopefully things in the DDR5 hardware department have settled down by then.

For me it will depend on its pricing.

I did a 3800X for $440 in Dec 2019
Then picked up the 5800X for $650 Jan 2021
Sold the 3800X for $400

So cost to me for that upgrade was basically $250. However with ADL-S now in the market I will not get the same return selling 5800X. I think used at best $300. So if pricing is reasonable on the 5800X3D I may jump. However if they price it at like $700 CAD I probably won't.

 

[nicalandia]

For me it will depend on its pricing.

I did a 3800X for $440 in Dec 2019
Then picked up the 5800X for $650 Jan 2021
Sold the 3800X for $400

So cost to me for that upgrade was basically $250. However with ADL-S now in the market I will not get the same return selling 5800X. I think used at best $300. So if pricing is reasonable on the 5800X3D I may jump. However if they price it at like $700 CAD I probably won't.

It will depend on what you do, if it's gaming all you need then 5800X3D will be the Gaming King until Zen4 comes in and lay waste to anything else(they demoed a all core 5 Ghz ES CPU)

 

[Joe NYC]

I would do side grade from a 5800X if the resale value is enough. That boost in performance could mean skipping Zen 4 and going Zen 5.

If the net cost of the "side grade" is within $200, I would consider it as well.

 

[Makaveli]

It will depend on what you do, if it's gaming all you need then 5800X3D will be the Gaming King until Zen4 comes in and lay waste to anything else(they demoed a all core 5 Ghz ES CPU)

This is where 3rd party reviews will come in. The price and performance has to be worth it for me to side grade! My workflows are not 100% gaming but if it meant prolonging the use of the system alittle longer at the right price I will jump.

 

[nicalandia]

This is where 3rd party reviews will come in.

With the info already given(testing hardware set up and games) we can extrapolate the performance in games like Far cry 6

 

[Det0x]

Pretty dissipated by AMD's 2022 CES keynote.. Would have bought 5950x3d for 1000$ in a heartbeat. (was planning to buy a few so i could bin them myself)
Feel like i have maxed my current 5950x (asus hero 8) platform and want something new to play with, and my MSI x570s UNIFY X MAX was bought solely for this reason..

Missed opportunity Lisa Su

And before anyone says something, my overclocking and computer tweaking hobby is very cheap compared to my other hobbies. (HIFI and cars)

 

[CakeMonster]

^ Not ragging on your hobby since I do the same, but knowing Zen4 is coming up in less than a year, switching 5950X for a slightly boosted 5950* sounds like a chore. A 5800X or 5900X, sure, if there are multithread use cases that are noticeable.

 

[Abwx]

Pretty dissipated by AMD's 2022 CES keynote.. Would have bought 5950x3d for 1000$ in a heartbeat. (was planning to buy a few so i could bin them myself)
Feel like i have maxed my current 5950x (asus hero 8) platform and want something new to play with, and my MSI x570s UNIFY X MAX was bought solely for this reason..

Missed opportunity Lisa Su

*edit*
And before anyone says something, my overclocking and computer tweaking hobby is very cheap compared to my other hobbies. (HIFI and cars)

Now that i think about it i think that there was a technical limitation and that this stacked cache is relevant mainly for servers, hence why there could be no 12/16C parts, but i m perhaps wrong.

The fact that the single core frequency is lower mean that diffusion of the heat is too slow because of the thermal resistance due to the stacked cache, because of this even a 12C with 6 cores per chiplet is not possible since the hotspot created by a core using 20W wont be better diffused.

This imply that such 12C/16C SKUs would be limited to 4.5GHz in single core, wich would be an unacceptable regression.